Palm vein recognition in few-shot learning via modified Siamese Network

As an emerging biometric technology, palm vein recognition shows many advantages that other methods do not have. With the continuous research in recent years, the implementation of palm vein recognition has developed from traditional methods to deep learning-based methods. However, deep learning requires huge numbers of labeled samples for training, and the palm vein datasets available for research are limited, which restricts the application of deep learning methods in palm vein recognition. This paper proposes a palm vein recognition method based on Siamese Network, which uses modified MobileNet V3 Large as the backbone network, it can achieve palm vein training and verification in few-shot learning. The method was tested on the Tongji University Palm Vein Database and the Hong Kong Polytechnic University Multispectral Palm Vein Database. The results suggest that the method is effective since it has 98.8% accuracy, 100% recall, 97.5% precision, 97.7% specificity , and 98.7% F1-score.